The Many Faces Fighting Disinformation - SAFEGUARDING CIVIL SOCIETY'S ROLE IN THE RESPONSE TO INFORMATION - EU DisinfoLab
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
The Many
Faces Fighting
Disinformation
SAFEGUARDING
CIVIL SOCIETY'S
ROLE IN THE Funded by:
RESPONSE TO
INFORMATION
DISORDERS
Table of
Contents
5 20
Global Disinformation Index
EDITOR’S NOTE “OUR GOAL IS TO DEFUND
DISINFORMATION”
VerificaRTVE
6 META-DEBUNKING
DECENTRALISED
DISINFORMATION
21
Tracking Exposed
KNOW YOUR ALGORITHM
22
Debunk EU Lie Detectors
8 DISINFORMATION IS SMALL
WATER DROPS THAT OVER
TIME CAN HEW OUT A STONE
NEWS LITERACY FOR ALL
AlgoTransparency
“ALGORITHMS NOT
10 24
Maldita.es DESIGNED TO HAVE QUALITY
JOURNALISM NOT TO BE INFORMATION IN THEIR
FOOLED OBJECTIVE FUNCTIONS
WILL NATURALLY FAVOUR
Who Targets Me DISINFORMATION
12 “WE USE TECHNOLOGY
26
TO TELL THE STORY OF First Draft
TECHNOLOGY” ADRESSING THE
INFORMATION ECOSYSTEM
14
MEMO 98 DISORDER
MONITORING ELECTIONS IN
THE DIGITAL AGE
Account Analysis
Citizen D
29 “MACHINE LEARNING OFTEN
16
GETS IT WRONG, I WANTED
COUNTERING
TO GO A STEP BACKWARDS”
DISINFORMATION FROM A
TO Z
30
2020, A TURNING POINT
17
WTFake!? IN OUR RESPONSE TO
THE FIRST CITIZENS’ ANTI- DISINFORMATION?
FAKE NEWS BRIGADE
18
Bellingcat
A CENTRAL NODE IN A
GROWING NETWORK
Key trends Timeline
1998 MEMO 98
2014 Bellingcat
These are micro-entities. 43% of the
initiatives surveyed count between 0 and
2 employees 2015 First Draft
Citizen D
2016 VerificaRTVE
They are community-reliant. 57%
work with volunteers and 64% rely on
crowdsourcing to some extent
Tracking
Exposed
Lie Detectors
Strained relationship with the private
sector. Almost a third of the participants
2017 Debunk EU
feel they are in opposition to the major
platforms, and two thirds have a weak
relationship with telecommunications
actors AlgoTransparency
WTFake?
Sustainability. Only one of the initiatives Who Targets Me
surveyed said that their operation is fully
sustainable.
2018 Maldita.es
Global
Dsinformation Index
Security. Cybersecurity is a concern for Account Analysis
all actors interviewed. None feels that
their operation is entirely secure.
3
ait
Clickb
N
RMATIO
DISINFO
4
As we have noted at EU DisinfoLab, more of. They exhibit new types of
disinformation has many faces expertise – from digital forensics,
(manifestations, motives, and tools). to crowdsourced research, to
It is only logical that the response journalism-classroom partnerships.
to disinformation must have many Though they might seem unusual
faces as well. In this project, we today in their techniques or
seek to present a panorama of the organisational structures, we believe
different kinds of actors responding that they also point towards the
to disinformation today - from future of civil society engagement
broadcast journalists to open source against disinformation.
investigators to election observers to
technology developers. In the report Despite the critical work they do,
that follows, we interview 14 actors many of these actors feel alone.
from across this emerging civil society They struggle to make their voices
ecosystem. heard by policymakers and by the
tech companies at the heart of our
It is important to note that many information ecosystem. Their long-
EdItor's NOTE
of the actors interviewed do not term security and sustainability are
see themselves as responding to not assured. In order to effectively
disinformation per se. We use the counter disinformation challenges,
term ‘disinformation’ as a shorthand we need an ambitious framework
for the many illnesses in our to sustain and further develop this
information ecosystem (what First civil society network. Following
Draft, whom we interview here, these interviews, we make several
refers to as “information disorders”). recommendations which we feel are
The problem for us encompasses necessary to safeguard a resilient,
disinformation and misinformation, decentralized civil society ecosystem.
mistrust in journalism and the
weaknesses or deficits in existing
media, the business model behind
clickbait and disinforming content, DESPITE THE CRITICAL WORK
algorithmic targeting, and other THEY DO, MANY OF THESE ACTORS
opaque mechanisms in our digital FEEL ALONE. THEY STRUGGLE TO
information architecture. MAKE THEIR VOICES HEARD BY
POLICYMAKERS AND BY THE TECH
This report is necessarily limited in
scope. The interviews here represent
COMPANIES AT THE HEART OF OUR
a snapshot or cross-section of an INFORMATION ECOSYSTEM.
expanding network of individuals
and initiatives. Still, the findings we Last, it should be noted that this
share are broadly representative of project is a view from the inside.
this space. We conducted qualitative, These discussions were framed by our
semi-structured interviews in order own experience at EU DisinfoLab, as
to understand how these initiatives a relatively young, small NGO, finding
were created and have evolved over our way in this new environment. We
time, and to examine in detail the consider the people featured here to
difficulties these actors face in terms be our colleagues, allies, and friends.
of sustainability, security, and impact. We admire them, we learn from them,
At the same time, these actors serve and we are continually grateful for
as examples of what we hope to see their work and their collaboration.
5
META-DEBUNKING
DECENTRALISED
DISINFORMATION
Redondo
Dr Myriam
Spain
The purpose of Verifica RTVE is to change the Global
culture of the institution.
D
r. Myriam Redondo has pioneered Steering Clear of Amplification
digital verification workshops for
Spanish journalists since 2012. She RTVE faces unique challenges
came into contact with RTVE as as a public institution formally
an external trainer in 2016. An early expert dependent on public
in the field, she released her doctoral thesis funds. Under heightened
on “Internet as a source of information for scrutiny from audiences
international journalism” in 2006. The team and inevitable political
she is currently part of, the digital verification pressure at moments,
team Verifica RTVE, includes only 2 full RTVE has to prove their
time and two half time employees (though independence day by
they will be adding new members in 2021). day and maintain the
Their purpose is to change the culture of public interest. In
the institution as a whole by teaching digital recent months,
verification. “This is transversal across RTVE this has meant
and that is the success. It involves documentary focusing more on
experts and archive experts, designers and public health
journalists, all positions in the team. We publish and less on
in multiple formats, radio television, internet...” fact checking
she explains. political
For Myriam, the problem of false news can
only be addressed transversally, through
journalistic capacity building and a cultural
shift. “We as journalists cannot do it all. If we
fight fake to fake, we’ll get tired. It’s like trying to
swim in a vast sea.” Instead she suggests the
need to meta-debunk, particularly for what
she calls distributed disinformation: “We
tend to analyse a fake, we take the content
and debunk it, but, at least in Spain, liars are
sophisticated in their activity. They publish
content but it doesn’t include a lie, it’s just a
suggestion, then a second liar goes farther,
and a third one farther. You have to debunk
the whole chain, the idea behind it”.
6 DIGITAL VERIFICATION FACT CHECKING JOURNALISM ACADEMIA
statements, which Myriam perceives very see what is happening. We are partially watching
often to be “noise”. “Politicians from extremist what happens in each but don’t have the whole
parties are tempted to use our services to amplify vision”. Myriam explains the need for more
a topic. When we verify a topic we enter their tools, in particular tools that provide network
agenda.” analysis and track trends across platforms. She
also needs the ability to parse more carefully
The team is also strategic in their method of between countries, to avoid unnecessary
debunking to avoid sharing content more transnational amplification through fact
widely than is necessary; they try to respond checking.
to queries from citizens in the same channels
where they are posed (directly in a WhatsApp Myriam wants
message, for instance) and they try to more collaboration
reply to personal between journalists
q u e s t i o n s and specialized
privately. Of institutions. She
course, this takes also sees a need for
IT’S IMPORTANT
massive human guaranteeing diversity TO DEBUNK FAKE
resources. Not in the growing industry BY FAKE BUT ALSO
only is this kind of digital verification; TO META-DEBUNK
of monitoring political fact checking TRENDS AND
t i m e is a clear example INTENTIONS
consuming, of an area where a
but it is often multiplicity of voices is
not possible needed, rather than a monopoly. “I envision a
in closed world in which journalists are doing our job, but in
messaging which we need [digital verification] organisations
spaces. for deeper analysis on a given trend of topic. Also,
“ W e a world in which citizens receive a more robust
cannot education on media literacy and critical thought,”
clearly she concludes.
About RTVE
RTVE is the Spanish Radio and Television
Corporation, founded in 1973. It is also
the first Spanish national media that began training
its staff on UGC (user generated content) and digital
verification techniques, at a time when political fact
checking was the dominant trend in the country. This
was before the election of Donald Trump, “the event
that changed everything”.
RESEARCH INTERNATIONAL RELATIONS / POLITICS 7
DISINFORMATION IS
SMALL WATER DROPS
that over time can hew out
a stone
as
Viktoras Daukš
DISINFORMATION ANALYSIS
Lithuania
MONITORING & REPORTING
ia,
Lithuania, Latv
Pol an d, US,
TECHNOLOGY DEVELOPMENT Estonia,
ac ed on ia
North M
STRATEGIC COMMUNICATION
LARGE-SCALE MEDIA LITERACY CAMPAIGNS
FACT CHECKING
COMMUNITY TRAINING
Debunk EU defines itself as an independent
technological think tank and analysis center
for disinformation.
8
F
or Debunk EU, building a response to have drawn on their expertise in research and
match the scale of the disinformation analysis, accumulated over many years.
problem has meant merging automation Currently they see potential for working with
and artificial intelligence with human Eastern partnership countries as well. “The
analysis and dedicated volunteers (a network technology is scalable, and we can work with local
of ‘elves’). “How do you find the needle in the partners”, say Viktoras. The fact that the Debunk
haystack, the needle being disinformation cases EU team in Poland was able to produce their first
that have the biggest impact? If you do it manually, report within four weeks of its establishment
you’ll never see the bigger picture”, explains in Warsaw proves that the system is not only
Viktoras. Debunk EU has developed an AI- scalable, but also able to achieve significant
based analytics tool which spots and identifies results in a short
topics of interest in online articles in real time.
time. This means that from 1 million pieces of
content they receive each month, their analysts D e v e l o p i n g
can focus on the most harmful ones (10 – 15 a u t o m a t e d IF YOU ARE A
000 content pieces). Long term reporting and solutions is costly BUSINESS, YOU
analysis are also at the core of their approach: and requires in- CAN BE FUNDED
disinformation analysts in the four countries house technical BY ANYONE, BUT IF
provide thematic reports on topics and trends, expertise, which YOU ARE FIGHTING
which are then shared with a wide range of is rare in the
stakeholders. Debunk EU has applied process disinformation
DISINFORMATION,
automation across their reporting and analysis space, and generally YOU CAN WORK
activities, allowing them to produce around 10 unheard of among ONLY WITH
reports per month. NGOs of this size. CREDIBLE AND
Debunk EU is TRANSPARENT
A scalable approach growing rapidly, and DONORS
now sustainability is
Founded in 2017, the project emerged in the the main question. As a tech-based organisation
Baltic countries and was supported by DELFI whose infrastructure is regularly attacked,
and the Google News Initiative. Debunk EU’s they must invest heavily in cybersecurity and
initial partnership with colleagues in Latvia monitoring of their digital ecosystem. Even the
and Estonia was inspired by their shared most agile project management requires fuel
history and disinformation threat constantly to run on. While they haven’t found the optimal
coming out of Russia since the fall of the Soviet financial model yet, Viktoras says that they will
Union. In order to meet this challenge, they be testing out new models in 2021.
Disinformation
• Coordinated efforts
• Trained forces Debunking
• Strategy in place • Fragmented efforts
• Well-funded and cheap • Difficult to work in real
to produce time
• Time consuming
• Reaching citizens is
expensive
9
Cost
JOURNALISM FACT CHECKING
NOT TO BE JOURNALISM
FOOLED
TRAINING
ADVOCACY
POLICY/EXPERT ADVISORY ROLE
Maldita is a Spanish non-profit fact-checking and data journalism
platform working to monitor and counter digital discourse, to fight
disinformation, and to promote media literacy.
T
wo Spanish broadcast journalists, chatbot, which currently has an answering
Clara Jiménez Cruz and Julio rate of 8 hours. Carlos is confident that the
Montes, began the organisation in AI will improve further as the community
2014. It has since grown to 20 full diversifies. Growing and diversifying the
time employees and between 30 and 35 community is top of the agenda, bringing
operating staff. The team primarily debunks in what he describes as “the many people
stories brought to them by their volunteer who care about disinformation but who aren’t
community and recirculates these debunks following every second of the news cycle.”
as widely as possible. Their approach is Maldita hopes to reach these wider audiences
deeply community centered, driven by a through new partnerships, for example with
tiered volunteer participation model: as of the popular tabloid 20 minutos. Meanwhile
July 2020, this counted about 40,000 General they’re exploring other verticals, for example
Malditos or recipients of the newsletter, through a science public engagement
1,000 Ambassadors or financial supporters, project, a platform focused on gender-
and 2,000 active Malditos and Malditas, related hoaxes, and a browser extension to
individuals who contribute ‘superpowers’ or promote transparency in public and private
subject area expertise to the organisation’s institutions. Recently, they are moving more
fact checking efforts. into the public policy space to weigh in on
major policy issues too big to ignore.
Since their founding and particularly during
the pandemic, the mis and disinformation Maldita brings clear benefits to many social
problem has changed in scale and scope. media platforms. In a sense, Carlos explains,
“We’ve gone from 1.5 million unique visitors per they “pay the bill” for fact checking services
month to sometimes 8.5 during the pandemic, that platforms claim to provide.
which is good and bad” Carlos testifies. “We
have been hiring during the pandemic, expanding Though Carlos specifies that different
our operation because the situation called for platforms have different practices, in
it. The audience was more engaged, but small general, these relationships are far from
organisations like ours are fragile. The scale of reciprocal. At the same time, platforms are
the operation was sometimes overwhelmed.” not sufficiently forthcoming with the data
that is necessary for actors like Maldita to
Maldita has been experimenting with assess the effectiveness of their debunking.
artificial intelligence (AI) in the form of a Still, “the task at hand is too important to
10forgo” in Carlos’ words, and
the community is in
agreement. Maldita has just
ndez
received foundation status Carlos Herná
in Spain, which allows them
to preserve their reputation Spain
and independence - a long
and expensive endeavor that
d
was made possible through Spain, North an
ic a
crowdfunding contributions. South Amer
WE NEED TO FIND
WAYS TO MAKE
SURE WE AREN’T
WORKING FOR
FREE. WE HAVE
TO ENSURE OUR
FUTURE
Why “Maldita”?
Maldita translates to ‘damned’. The name is a
reference to Maldita’s first fact checking initiative,
Maldita Hemeroteca (Damned Archive), which is a
project to confront politicians with past statements
they themselves made that contradict their current
views. 11“WE USE
TECHNOLOGY Sam Jeffers
TO TELL THE STORY OF
om
United Kingd
Global
TECHNOLOGY”
CONSUMER LITERACY RESEARCH TECHNOLOGY DEVELOPMENT POLICY/EXPERT ADVISORY ROLE
ADVOCACY
Sam Jeffers co-founded Who Targets Me in 2017, having worked
previously on political campaigns which had favoured grassroots,
bottom-up tactics. He’d witnessed the ecosystem shift around 2015,
when parties and candidates began to buy social media ads to target
small groups of voters with large quantities of tailored messaging.
W
ho Targets Me is interested platforms. They are a small operation and all of
in “practical transparency for their funding is project based. Sam and his co-
political campaigning”, in Sam’s founder, who has another fulltime job, take on
words, and most focused on capacity and particular skills as needed, which
transparency for people running for office. gives them a certain lightness compared to
“Others might be traditional NGOs. “We’re a bunch of flexible
looking for people who have interesting and innovative ideas,
other things which we try to execute cheaply and quickly and
such as fake simply.” Despite their size, they are committed
WHILE WE DO profiles or to staying “at the forefront of thinking on
COMPLAIN ABOUT state influence political ads and regulation”. They
THE PLATFORMS (A efforts. We’re bring a more balanced voice to
LOT), SOMETIMES looking at the the policy space, where new
THEY’RE THE very top of the entrants are often eager
tree because to simply “see everything
ONLY ONES DOING
we think that banned”.
ANYTHING sets the norms
and examples A lack of creativity in
around which the rest of politics and democracy the policy response
work.” implies a general need
to better bridge the
Who Targets Me is working in an uncrowded product-policy divide,
space, in part due to the technical challenge, Sam thinks. “Two years
and in part due to an increasingly chilly ago we could have been
research climate around the large social media experimenting with labeling
12in an independent way and looking through a more shaped over a longer period of time. We should
imaginative portfolio of responses. Alternatives be looking for longer and using the aftermath of
are possible, but no-one is really doing the work elections to push forward with reforms.”
of designing what better services might actually
look like.” He also notes the need for longer
term thinking. “We need to think about election
years and cycles rather than the few months and
weeks before elections. People’s opinions are
Consumer ad literacy
Who Targets Me provides a free browser extension
to help people understand more about the paid media
they are exposed to on Facebook. The software shows
a library of all political ads sent to them and data on
who is showing them the most ads, and also helps
them understand the mechanics behind the tailoring
of that content. Sam describes it as a consumer ad
literacy tool. “We’re trying to exemplify the transparency
that we want the platforms to provide.”
“WE WANT TO UNDERSTAND
WHAT THE BIG ACTORS ARE
DOING, WHO’S FUNDING
THEM, AND HOW THAT HAS
AN IMPACT ON DEMOCRACY
AND ON PEOPLES’ UNDER-
STANDING OF ISSUES.”
13MONITORING ELECTIONS IN THE
DIGITAL AGE
MEMO 98 is a monitoring organisation consisting of media
and election experts. What began in 1998 as a project to
monitor the Slovak media prior to the parliamentary elections
developed into a permanent organisation that has conducted
media and election monitoring across the world.
MEDIA MONITORING
ELECTION OBSERVATION
SOCIAL MEDIA MONITORING
Elections toolkit
JOURNALISM
In November, Rast’o published a toolkit “How
CAPACITY BUILDING / TRAINING to monitor media coverage of elections”. The
toolkit was developed within the context of
POLICY / EXPERT ADVISORY ROLE the Council of Europe project “Supporting the
transparency, inclusiveness, and integrity
of electoral practice in Ukraine”,
C
urrently MEMO has a core implemented within the framework
team of 7 full time employees, of the Council of Europe Action
but they work closely with Plan for Ukraine 2018–2022.
local partners in a number of
countries. MEMO’s methodology of
media monitoring focuses on content
WE AIM TO HIGHLIGHT
THE WORK OF LOCAL
PARTNERS, SERVING
AS A SORT OF QUALITY
GUARANTOR
and aims above all to evaluate political
and social diversity in media reporting.
In recent years, they’ve adapted their
methodology from traditional media to
account for the principles of the social
web, but they maintain the focus on
14both spheres. “This combination of traditional be the ones left after international election
and social media monitoring is very important,” observers leave. “It doesn’t end at the end of an
Rast’o explains. election cycle. It is usually the start of another
one.”
Their monitoring is designed to provide in-
depth feedback on pluralism and diversity Where MEMO struggles, along with others
in media reporting, including coverage of in this space, is in fully understanding the
particular themes (integration of minorities, mechanics of amplification online. “We focus
corruption etc.). MEMO does not only focus less on inauthentic behavior - bots and trolls who
on disinformation. They have “a more holistic amplify content. But from a different perspective,
and general approach, assessing both the positive
this is a critical part. This can make marginal voices
and negative impacts of social media platforms more visible.” The tools currently available for
on election integrity” in Rast’o’s words. They social media analysis (for instance, Facebook’s
study three things: the actors (from both CrowdTangle) have been a game changer for
traditional media and social media), the MEMO, but the data is incomplete without
messages and narratives (how they are used access to private pages and closed messaging
by parties to make claims and to polarize) and spaces. While entering these kinds of spaces
the messaging (how the message is amplified). raises ethical questions for researchers, there
are legitimate design and transparency issues
Much of MEMO’s work includes capacity that hinder election integrity monitoring.
building and training, enhancing the media
monitoring activities of local partners. They
also train journalists, NGOs, regulators, and
other members of the media. For Rast’o, it is
critical to support these local actors who will
Rast’o Kužel
Slovakia
Global
PLATFORMS SHOULD BE
GEOGRAPHICALLY BLIND
WHEN IT COMES TO
APPLYING THEIR RULES.
WE SHOULD NOT ASSUME
THAT EVERY ELECTION IS
LIKE THE ONE IN THE US
15Citizen D’s COUNTERING MISINFORMATION
FROM A TO Z
ongoing
campaigns
Citizen D is currently
tracking two governmental Based in Ljubljana, Slovenia, Citizen D
ad campaigns which are is a nonprofit whose core mission is the
being funded with tax-
payer money. The web and promotion of human and digital rights.
D
television ads are directly
espite counting recommendations.
related to government
only two full time
parties through various non-
employees, Citizen D’s In the Slovenian political
governmental organisations
activities are various context - which Domen has
with strong government ties.
and in-depth. Their actions range written about - Citizen D faces
Citizen D is relying mainly on
from privacy rights monitoring a double battle. In addition to
FOIA requests to gain insight
and training, to investigating the deeper problem of political
into the government’s
practices within the advertising propaganda funded by public
opaque process of cost
industry and holding discussions money, they have to respond to
tracking and selection of the
on the purpose of mass media a prevailing narrative around
ad placements.
in a democratic society. They so-called ‘fake news’. They have
offer media literacy and digital developed a targeted approach,
privacy trainings, and they raise following the money between
INVESTIGATIVE REPORTING
awareness through campaigns. all relevant public services
MEDIA LITERACY
One example is a national and actors (advertisers, public
anti-hate campaign drawing funds, political parties, etc.).
CAMPAIGNING attention to the Slovenian This posture can make financing
public funding of Hungarian complicated. Citizen D tries to
propaganda outlets. maintain a dispersed model of
funding, developing alternative
While Citizen D aims to sources of revenue that include
encourage active citizenship commercial projects - in line
and democratic participation, with their mission and under a
Domen Savic they take legal action on their code of conduct. Meanwhile,
own. “This myth around an active the lack of outside, international
Slovenia citizen that will react and gather media pressure on Slovenia
sufficient information from the means that local issues are often
media to make their own case, ignored. “I’d have an easier time
Balkans
that’s not the reality we’re seeing,” saying there was a problem with
Domen explains. To help lift Russian propaganda. Nobody is
the onus from citizens, they’ve focused on Orban and Vishegrad,”
adopted an approach of working says Domen.
from A to Z. “Our job doesn’t
THE TERM ‘FAKE end when we file a report. From
NEWS’ HAS IN OUR that we analyze the problems,
OPINION MORPHED we define which decision makers
INTO A CATCH-ALL are responsible, and we pursue
PHRASE THAT DOES that.” Frustrated by the limited
MORE HARM THAN functioning of the legislation in
some areas, for example around
GOOD false advertising, they have also
16 taken to proposing legislativeTHE FIRST CITIZENS’
ANTI-FAKE NEWS BRIGADE
“You could say the problem I’m trying to solve is fake news,
but, deeper than that, the problem is mistrust of journalists.”
JOURNALISM ACTIVISM
A
ude is a journalist by training. She is have become discussion moderators. Others
also president of Fake Off, a media contribute in different ways, like developing
literacy association that works with the logo. Her colleague Sylvain Louvet helps
young people in classrooms, which with editorial and video production. Aude
she founded together with a small group now has 1411 people on her Discord, and the
of journalists in the wake of the 2015 Paris feedback from many of her followers suggests
attacks. Behind the crisis of media literacy and she is achieving the impact she hoped for,
‘fake news’, Aude identifies a lack of dialogue regularly receiving comments like “usually I
and a growing mistrust between journalists don’t like journalists, but I like you!”
and people. Following the election of Donald
Trump, she began to feel that it wasn’t sufficient “C’est le bazar”
to intervene in classrooms, and decided to
take directly to YouTube. “The motivation Organisationally, things are messy, Aude
behind everything is anger, I see the impunity, the admits. She set out on this alone, without
indecency of manipulating the public, it makes me a financial model, and without the kinds of
so angry.” contacts that many need in this space to
survive. A few crowdfunding links have not
La rédac’ WTFake (‘the citizen journal’) tracks yielded much yet. “I’m just a little investigative
and debunks conspiracy theories, primarily journalist, but now I find myself having to reflect
found on YouTube, with the support and like a business person” she explains. While
admiration of an online following. In a way, the energy of her followers motivates her to
Aude’s activities mirror those of her enemies continue, animating the community also limits
- like the conspiracy theorist Jean-Jacques her ability to strategize and grow. To devote
Crevecoeur - in a multimedia chase that plays time to fundraising would mean abandoning
out across the social web. The investigations the community, and both activities take away
progress over several days on a Discord chat, from time spent on investigations. Meanwhile,
and culminate in a revelation streamed live there are trolls to be wary of. Aude has already
over Twitch. A handful of dedicated followers changed her phone number after a conspiracy
theorist doxxed her online. “Given the people I
want to investigate, I’ll probably see more of this,”
she reflects.
WTFake!?
Aude Favre
France
France
17Eliot Higgins
Netherlands
A CENTRAL NODE IN A Global
GROWING NETWORK
Bellingcat emerged in July of 2014 from the online community
that had formed around Eliot’s blog, in particular around his work
tracking the downing of flight MH17.
OSINT / DIGITAL FORENSICS JOURNALISM JUSTICE & ACCOUNTABILITY MEDIA & REPORTING TRAINING
I
launched Bellingcat wanting to give verification component. “You’re
people a place to publish articles often dealing with debunking one side
about what they were doing using through open sources,” says Eliot.
open source, and also to create Their work revolves around three
resources for people to learn how to do it steps. First, they identify information
themselves.” Today, Bellingcat is a fully online related to a topic of interest
fledged organisation with 18 core or importance (subjects include
staff across research, business and corruption, corporate misconduct,
administration, a management team, racial equality, far right movements,
a supervisory board, and “with proper etc., but it is important that interests
policies” to quote Eliot. It has recently be led by the team). Next they verify
been registered in the Netherlands as that information using open source
a Public Benefit Corporation. Though techniques, drawing strategically on
it is no longer a volunteer dominated volunteers and on an engaged social
organisation, Bellingcat has retained media community who can help with
the volunteer community as a key precise aspects like geolocation and
element: a network of open source identification. Finally they amplify
investigators spread the use of that story, through a report, video,
open source methods while further or podcast, or even as courtroom
developing tools and methodologies, evidence. Bellingcat is increasingly
maintaining a focus on justice and exploring the role of open source
accountability. evidence in legal processes; they have
worked already with the ICC and the
Identify, verify, amplify United Nations.
Bellingcat is not in the business of fact Much of what Bellingcat does could
checking so much as “fact finding”. be considered capacity building.
Still, disinformation is inherent to They offer training to journalists and
their activities, in that online open fact checkers as well as to activists,
source investigation has a strong NGOs and lawyers. These trainings
18also provide a source of revenue (currently 30 collaboration, we’ll build coalitions of groups
percent of their funding). Eliot sees Bellingcat to work on topics.” They recognize this can
as “a central node in the network that makes up be strange for media organisations who are
the online open source investigation community”, accustomed to scoops and exclusives, so
a universe which spans human rights they’re strategic in bringing together non-
organisations, major media competitive media and in building trust.
outlets, and individual
Twitter users. Though many organisations look to Bellingcat
Collaboration is key. as an example, Eliot readily admits that
“Often when we come “we’re figuring this stuff out as we go”. Beyond
across a project OSINT, the team is focused on the editorial
or something to and production side; Bellingcat is training
investigate, and their researchers in journalistic writing and
if it goes beyond building up a production company (which
the scope of would be another source of income). They’re
what Bellingcat also leaning into their volunteer community
does or if it is (through a volunteer app and a Patreon),
something and trying to increase their impact with
that can international governance stakeholders, for
benefit instance through developing standards for
from a open source evidence in the area of justice
and accountability.
Black lives matter
Along with Forensic Architecture, Bellingcat
geolocated and verified over a thousand incidents
of police violence during the Black Lives Matter
protests in the United States. The project analyses
them according to multiple categories, and presents
the data in an interactive cartographic platform.
19“OUR GOAL IS TO DEFUND
DISINFORMATION”
The Global Disinformation Index is responding to the growth of
digital disinformation sustained through advertising revenue.
BRAND SAFETY TECHNOLOGY DEVELOPMENT RESEARCH & INTELLIGENCE ANALYSIS POLICY / EXPERT ADVISORY ROLE
T
he Global Disinformation Index is a not- approach provides a disinformation risk score for
for-profit organisation founded in 2018. websites which advertisers can use in real time.
Registered in the United Kingdom, they
have a virtual team of 15 experts based GDI has already seen impact, for example, in
around the world. “We focus on the advertising Google’s decision to defund all Coronavirus
networks that place ads on websites regardless of conspiracies. But the change isn’t happening fast
whether the advertiser knows or wants their ads to enough, and it isn’t systemic. “They’ve [Google]
end up there,” Clare explains. GDI’s known about anti-vax conspiracies for years, but
core contribution is to provide the virus was 7 or 8 months old before they made
the advertising industry with that decision. What about all the other anti-science
disinformtion risk ratings conspiracy theories out there affecting our ability
to assess whether or not to think critically? Flat earth conspiracies, climate
ads should be placed change denial...” Clare reflects. Moreover, other
on certain sites, but tech companies providing ad services must
they support other take the same actions for a whole-of-industry
stakeholders as well, response, not just Google.
including through
their media market A young organisation, GDI needs to ensure
risk rating reports. their own sustainability, primarily by building
out new commercial products and services to
GDI views its effort fund its not-for-profit work. Looking forward,
as aligned with “Brand they hope to solidify “a coalition of the willing”
Safety” efforts, except that against lucrative disinformation, aligning with
the organisation is focused on the corporate responsibility agenda. They also
disinformattion, an area that was want to join forces with other sectors to expose
previously ignored. They brought with other distortions hat provide a funding lifeline to
them a degree of insights and intelligence that was disinformation. “Payment systems, merchandising,
previously lacking. “Brand safety usually involves e-commerce… there’s a whole ecosystem and that is
keyword blocking, static lists... Until GDI came along, not well enough known,” Clare concludes.
there has never been a box for highly disinformative,
toxic, adversarial narratives,” says Clare. GDI
makes use of a comprehensive framework, both
human and AI powered (humans can’t assess the
whole internet at “colossal speed and scale”, while Clare Melford
AI isn’t nuanced enough to differentiate high
production, high traffic media outlets that publish
UK, Germany
disinformation). This combined, innovative
Global
20i
Claudio Agost
KNOW YOUR Netherlands
ALGORITHM Netherlands,
Italy, Argentina
Tracking Exposed is a non-profit, free software project that
aims to analyze evidence of algorithmic personalization. It
was founded in 2016 by Claudio Agosti, a self-taught hacker
and developer, currently researcher at the University of
Amsterdam.
RESEARCH ACADEMIA DATA ANALYSIS TECHNOLOGY DEVELOPMENT ADVOCACY USER EMPOWERMENT
I
n 2018, the University of Amsterdam’s onboarding, potential damage and security
Department of Media Studies expanded risks.
on Claudio’s work, creating the project Claudio is working to make the tool simpler,
ALEX - Algorithms Exposed, Investigating with more visual results. He would like users
Automated Personalization and Filtering for to be able to play with and control their own
Research and Activism. Through his work, algorithms. He’s also interested in pursuing
Claudio empowers end users to understand strategic litigation using his findings as
how aggressively the algorithm is mediating evidence of platform rights violations. He
information for them. The project has been has done election-related monitoring in
applied to four platforms so far. “For Facebook the past, and he has a project coming up to
and YouTube, it’s about information quality. For work with a polling company. Elections are
Amazon and Pornhub, it’s about explaining how emerging increasingly a business case, he
algorithms exist in other places and can still have observes; offering this high expertise service
an impact on you.” could be a form of sustainability for Tracking
Exposed. However, advertising composes a
Claudio mostly works with researchers in small portion of our informative experience,
the Netherlands and in Italy, though he has and so we should remain focused rather
also carried out work related to Argentina. on the role of our personalized algorithms,
The academic label is important for the Claudio notes: “We
legitimacy of the project, and also for finding should have control
collaborators and growing the project. of our algorithm
The work on Pornhub was carried out by a because that is the
researcher he met while teaching at the Digital tool responsible
Methods Summer School, who wanted to for all the content
expose heteronormativity on the platform. selected.”
Tracking Exposed gathers data through
crowdsourcing, but getting access to wider
audiences and groups is challenging. To the
extent that outreach depends on marketing
and visibility, this can exceed the capacity
of a researcher. For a small programming
project like Claudio’s, recruitment requires
compromises and calculations: investment in
21NEWS LITERACY
Juliane von
Reppert-Bismarck
FOR ALL
Belgium
ia,
Belgium, Austr
Germany
Juliane founded Lie Detectors to respond to two sides of a
problem she identified in the information ecosystem.
JOURNALISM EDUCATION MEDIA LITERACY POLICY / EXPERT ADVISORY ROLE
O
n the one hand, she was it is possible because Lie Detectors has
dismayed by increasing distrust that rarest of gifts in this ecosystem: long
in professional journalism, term flexible funding). From a research and
increasing polarisation, and the methodological perspective, this diversity
blurring of facts and opinions - a problem to is crucial. The organisation also works
which children are particularly vulnerable. closely with questionnaires and feedback
And on the other hand, she noticed “the forms to identify developments in mis and
succumbing of journalism to the promise disinformation from the perspective of the
of click bait”, the response of an industry in children. Juliane explains that conspiracy
crisis forced to take short cuts. theories, which appear to be a growing
challenge for students, will require particular
Lie Detectors brings the two sides together, policy recommendations and particular
placing working journalists into schools to training for educators: “Conspiracy theories
deliver dynamic trainings. The organisation are the ultimate deep fake, because they are
currently has a core team of 14 staff and layers upon layers upon layers that you have to
coordinators, along with a network of 200 unravel.”
journalists and a growing ecosystem of
educators and schools in Belgium, Germany “This is not something we should be feeling
and Austria. The approach is circular: Lie proprietorial about”
Detectors trains journalists, who then train
children and educators in the classroom; The service Lie Detectors is providing is in
meanwhile that classroom experience demand, both on the side of schools and from
provides feedback for journalists and journalists (they currently have a wait list).
newsrooms, as well as insights for Lie “This is a method and a structure. It’s so simple,
Detectors’ policy development and advocacy it makes perfect sense… We want to amplify
work. and facilitate organisations doing similar
work.” They increasingly hold seminars that
Sessions are free and journalists participate enable teachers to deliver these trainings
as volunteers with compensation, which themselves.
means that trainings are delivered in diverse
school environments and from journalists
with diverse experiences. (Incidentally,
22Thinking critically
Formerly a journalist herself, Juliane understood the
need to “differentiate ourselves from disinformation,
and also admit that we don’t always get it right.” She
explains that “admission of fallibility and rebuilding
of trust was important from the beginning. That’s why
we have to work with a particular kind of journalist,
capable of speaking critically about their work.”
EVERYBODY KNOWS
DISINFORMATION IS A
PROBLEM, BUT PEOPLE
NEED TO UNDERSTAND
THE CAUSES AS WELL AS
THE SYMPTOMS. THAT
BECOMES MORE VISIBLE
WHEN YOU ARE ABLE TO
MAKE A COMPARISON
BETWEEN COUNTRIES,
LANGUAGES, CULTURES,
SCHOOLS.
23“ALGORITHMS NOT DESIGNED
TO HAVE QUALITY AlgoTransparency
INFORMATION IN Guillaume Chaslo
t
THEIR OBJECTIVE FUNCTIONS
France
Global
WILL NATURALLY FAVOUR
DISINFORMATION”
RESEARCH ARTIFICIAL INTELLIGENCE PROGRAMMING ADVOCACY
Guillaume Chaslot is a computer programmer with a PhD
in artificial intelligence, the founder of the consulting firm
IntuitiveAI and of the nonprofit AlgoTransparency. He is currently
a Mozilla Fellow. During his three years at Google, he had worked
beside YouTube engineers on their recommendation system. He
observed how the algorithm, which optimized for watch time,
had some dangerous side effects.
G
uillaume particularly became aware the problem - he left the company and began
of the “snowball effect” boosting AlgoTransparency in 2017, with the objective
conspiracy theories on the platform. of exposing what content recommendation
“I realized the algorithm was promoting algorithms are showing people on different
disinformation, very often more disinformation platforms. They currently look at Facebook,
than truth. I looked at topics which were clearly Google Autocomplete, Twitter Trending, and of
disinformation, like flat earth theories, and realized course YouTube, where they monitor over 800
that the algorithm was producing more flat earth top information channels. The project runs on
than round earth videos.” Guillaume small grants and is now supported by
had identified a fundamental Guillaume’s Mozilla fellowship.
design flaw playing out at a
massive scale. “We can only At the moment,
do so much fact checking. A l g o Tr a n s p a r e n c y
If the algorithm decides works primarily with
70% of the views, it’s a journalists. The
losing battle.” functioning of the tool
is a bit complicated,
Unable to convince and it takes time to
his colleagues to explain how it works.
intervene - even Guillaume would like
to recognize the tool to be useful
24YOUTUBE KEY PRODUCT AND POLICY
to more people and broader in what it LAUNCHES TO RAISE AUTHORITATIVE VOICES
surveys. In general, he still sees a huge TO REDUCE THE SPREAD OF BORDERLINE
need for transparency into algorithms. CONTENT SINCE 2015
“People either don’t know how to do it
because they weren’t insiders, or are afraid • Raise authoritative content
• Reduce spread of borderline content
because they don’t want to face big tech
companies, or discouraged because there
isn’t a real business in it” he says. Access to 2015
data is still an issue, and it is insufficient July 27
for platforms to be able to decide what Improved ranking system to reduce
the visibility of junk comments
information they share publicly. What’s
needed is external pressure.
2016
Ideally, Guillaume would like to build out
September 14
a team and expand monitoring across Launched YouTube Player for
a range of platforms, then be able to Publishers for the news industry
work with fact checkers, analysts, and
journalists to make this data digestible 2017
for different stakeholders: individuals,
researchers, and the employees at these July 20
tech companies themselves - many of August 18
Redirected users seeking violent
whom don’t want to look at their own extremist propaganda-related content
Launched Top News sheld in search
to playlists that debunk its mythology
problem. “Instead of small investigations, results and Breaking news sheld on
the homepage
I want to look at the general issue. I want
to make the data available and enable
everyone to use it as they want.” September 14
Started to surface authoritative
content in search results
Inspiring 2018 February 2
Launched information panels
change
alongside videos from government or
public-funded publishers to provide
July 9 content
• $25m investment to
AlgoTransparency has had enormous support trusted journalistic
impact. Guillaume himself has organisations and improve the
news experience on YouTube
achieved media recognition as a • Launched information panels
October 16
whistle blower of sorts (he recently providing context on topics
Launched information panels to
prone to misinformation and
appeared in the film “The Social conspiracy theories
provide context on US election
candidates. Later, for EU Parliament
Dilemma”), as the policy discussion has • Launched Top News sheld on
election candidates too
the YouTube homepage
increasingly turned towards the need
for algorithmic transparency. The
impact on platforms is more difficult January 25
2019
to understand. “YouTube changed their Reduced recommendations of
borderline content and videos that March 7
algorithm 30 times to address the exact could misinform people in harmful Launched information panels in India
issues I was talking about”, Guillaume ways in the US to provide fact checking in Youtube
search
notes, referring to their actions in June 3
June 5
2017 on the amplification of terrorist Improved our machine learning
Raised authoritative content in Watch
classifier to better apply our
content. But they did not take action protections for content featuring Next panel for people engaging
with borderline content in the US.
regarding disinformation minors and risky situations, across
Further reduced recommendations to
more videos
like the flat earth content borderline content in the US
that he had brought to July 8
Started reducing recommendations
their attention. to English-language borderline
videos outsie the USFirst Draft
Dr Claire Wardle,
Marie Bohner
alia
UK, US, Austr
Global
ADDRESSING THE
INFORMATION
ECOSYSTEM
DISORDER
26W
First Draft was founded in ith offices in three time zones
(London, New York, and Sydney),
2015 as a nonprofit coalition and a team of 52 employees,
of nine founding partners First Draft is monitoring around
the world and around the clock. First Draft is
with the mission of protecting addressing an “information ecosystem disorder”,
communities from harmful says Marie. This work is guided by their
Information Disorder Framework which Claire
misinformation. That original Wardle, co-founder and US director, first laid down
coalition has expanded greatly in a 2017 report for the Council of Europe with
co-author Hossein Derakhshan. This framework
since, and now includes helped put First Draft on the map, but it is one of
international partnerships many approaches the organisation has pioneered
since its founding. CrossCheck, a collaborative
with newsrooms, universities, approach to reporting around elections, has
companies (recently they have been used in the US, UK, France, Germany, Brazil,
begun work with Spotify), Nigeria and Spain, and inspired similar initiatives
in many other countries. “Elections are our DNA”,
institutions such as the WHO or Marie explains. First Draft has developed a strong
UNICEF and other NGOs. editorial department, which produces daily and
weekly briefings from their global monitoring
and they contribute regularly to the research
discussion. Through their global Partner Network
of newsrooms, fact checkers, human rights
INFORMATION DISORDER HEALTH organisations and technology platforms, they
have been working to share knowledge and set
JOURNALISM POLICY / EXPERT ADVISORY ROLE
standards in the areas of collaboration, training
RESEARCH GLOBAL COLLABORATIVE NETWORK
and research.
ACADEMIA CAPACITY BUILDING Rooted in strong local partnerships and
collaborations, First Draft takes a global
approach in order to gain a cross-
border view of information
The Deceptive seven: Common types of
misinformation & disinformation
Satire or parody Misleading content Imposter content Fabricated content
No itention to cause Misleading use of When genuine sources New content that
harm but has potential information to frame an are impersonnated 100% false made to
to fool issue or individual deceive and do harm
False connection False context Manipulated content
When headlines, visuals When genuine is When genuine
or captions don’t support shared with false information or imagery is
the content contextual information manipulated to deceive
27disorder. “This approach aims to overcome The organisation is growing quickly, but rapid
individual biases, to compare stories, and to share expansion during the pandemic has been a
similar problems,” Marie explains. challenge. While they have had plenty of work
to do around elections - particularly the US
election - they still lack long term, flexible
finance that guarantees an organisation
sustainability and peace of mind. “It’s
uncomfortable being 100% dependent on funding,
so we are developing diverse revenue streams to
WHILE THE MEDIA TEND TO ensure we can continue supporting those relied
FOCUS ALL ON THE SAME THING, upon for accurate information in the years to
WE TRY TO CONCENTRATE ON come”. Meanwhile, there is no lack of work to
WHAT PEOPLE ARE LOOKING FOR be done. While continuing their focus around
ANSWERS TO democracy and health, particularly vaccines,
they are already looking ahead to other global
topics that are vulnerable to disinformation,
such as the looming economic downturn and
Recently, First Draft has been exploring what the climate crisis. There is more awareness
they identify as data deficits, areas where a lack and acknowledgement of the threats and risks
of relevant and readily accessible information of online disinformation, an urgent need to
may lead to misinformation. They are trying empower societies with knowledge and skills,
to identify people’s questions, for example, and for collaborative efforts to increase. After
using Google Trends, or else through direct 5 years building experience and expertise,
conversation with their audiences. To better First Draft is needed now more than ever
understand misinformation and data deficits before.
in closed spaces, they apply crowd-sourcing
strategies, such as tip lines, where members
of the public flag content for examination.
Capacity
building Dr Claire Wardle
First Draft is building capacity across all imaginable
stakeholders. In addition to their series of trainings
for journalists (from personal and cyber security to
the ethics of investigations, etc.), their Local News
Fellowship Program around the US 2020 election
equipped local and regional journalists in social
media monitoring. They pioneered information
crisis simulations to help build resilience, and have
a growing number of resources targeted at more
general audiences as well.
Marie Bohner
28Account Analysis
“MACHINE LEARNING OFTEN GETS IT WRONG,
Luca Hammer
I WANTED TO GO Germany
A STEP BACKWARDS” Global
Luca began the project behind Account Analysis one weekend in
2017, in response to a growing debate about bots.
RESEARCH DATA ANALYSIS OSINT TECHNOLOGY DEVELOPMENT
J
ournalists were talking about social bots Luca is less interested in bots nowadays. “I’ve
who influence elections and debate online.” discovered bot-nets on Twitter and published
While he knew a handful of existing tools about them, but I think their impact is low.
to automatically identify bots on Twitter, The platforms have great tools and try to stop
usually by giving accounts a score, it was difficult spamming.” He’s more focused on platform
to understand how those scores were produced. manipulation. “It doesn’t matter if it’s one
The research community tends to agree that person or several people, I’m interested in what
most of the tools used to identify bots are flawed. they do and what they try to achieve.” Account
“It’s nearly impossible to reliably identify bots from Analysis helps him study this because it lets
the outside; platforms can do this better because him understand how those accounts work. He’s
they have access to more data points.” also interested in false information sharing in
big public channels on Telegram. In the future,
His approach has been to build a tool that wouldn’t he aims to make more tools on more platforms
deliver a score, but rather allow users to evaluate accessible to more people. He also has plans to
for themselves. Account Analysis lets users study add features, including tools to be able to analyse
Twitter accounts by looking at criteria like how debates, not simply accounts.
many retweets they post, which hashtags they
use, which websites they link to most often.
“For researchers to fully understand what’s going Keeping it free
on, I’m convinced that you have to see the data
Account Analysis has both a free and
yourself.” Luca believes it isn’t enough to just
paid version (for 15 euros a month, the
“get the numbers” of how often contents are
tool essentially works more quickly). He
shared. In order to understand influence and
developed this ‘freemium model’ after
impact, researchers need to be able to see the
finding the business case for marketers,
data directly. This requires more communication
who use the tool to evaluate the accounts
between platforms, the researcher and
of influencers and politicians. This business
programmer communities. Twitter’s API is by far
case covers his server costs and
the most accessible for researchers: anything
allows him to continue to
public on the platform is public on the API.
provide the tool for
Telegram, which Luca is beginning to focus on
free. His primary
© Tony Gigov
more closely, is similarly unrestricted. Facebook,
motivation is to
who’s API is notoriously difficult to use, has a less
serve the OSINT
clear cut user-facing definition of private and
and journalist
public space, which contributes to the challenge.
c o m m u n i t y.
Still, all platforms could do more to make their
He also offers
APIs accessible for research.
workshops for
journalists.2020, A TURNING POINT
IN OUR
RESPONSE
TO DISINFORMATION?
T
his research was conducted at a civil society ecosystem. Disinformation is
moment when the disinformation a horizontal challenge by which more and
challenge has never seemed higher. more actors find themselves confronted.
From the Covid-19 health crisis and A decentralized architecture can catalyze
parallel ‘infodemic’ to elections in the US and multiplier effects and build capacity and
Belarus, 2020 has been a tumultuous year for resilience among those currently at the
our information ecosystems. It has also been periphery of the disinformation threat (climate
a significant year for regulatory response to activists, health professionals, etc.).
disinformation. The European Democracy
Action Plan, a roadmap to enhance democratic Our aim in this project has been to try to
integrity and resilience across the European understand this growing civil society network
Union, was released in December. It was - their struggles and their successes - in
followed by the Digital Services Act and the order to better support their activities and
Digital Markets Act, regulatory packages to fortify this ecosystem as a whole. Far from a
clarify the role and responsibilities of online comprehensive survey, this research can only
platforms and increase accountability for scratch the surface of this vast landscape of
online disinformation. actors. And yet even with this partial view or
snapshot, we have been able to highlight trends
Regulation of online platforms will reduce and shared challenges, to pinpoint areas where
opacity and better equip anti-disinformation collaboration is possible and where more
actors like those interviewed here, but support is necessary. It is our hope that these
regulation is not a silver bullet. As a diffuse findings will further empower these actors and
and rapidly evolving set of challenges, those like them, and provide evidence for the
disinformation requires a broader response. competent authorities to act.
Key to this response is a thriving, decentralized
30You can also read
